Reduced Phase Space Formalism

• Reduced phase space formalism is a method that defines physical observables by restricting the full phase space to a constraint surface and then quotienting gauge orbits.
• It employs techniques such as Marsden–Weinstein reduction, Dirac constrained Hamiltonian methods, and cohomological quotients to handle gauge redundancies in mechanics, field theory, and gravity.
• The approach underpins reduced quantization and path-integral formulations while addressing singular quotients, boundary effects, and the distribution of symplectic and Poisson structures.

Reduced phase space formalism is the construction in which one starts from a kinematical or extended phase space, restricts to a constraint surface, and then quotients by the gauge orbits generated by first-class constraints. In finite-dimensional symplectic language this is the Marsden–Weinstein quotient

$\mathcal{M}^{\mathrm{red}}=\mu^{-1}(0)/G,$

while in Dirac’s constrained Hamiltonian language it is the passage from a constrained phase space with gauge redundancy to a phase space of gauge-inequivalent physical states endowed with the induced symplectic, Poisson, or Dirac-bracket structure (Lysov, 2021). Across mechanics, gauge theory, and gravity, the formalism provides the canonical arena for physical observables, clarifies the status of conserved quantities as Dirac observables, and supplies the starting point for reduced quantization, covariant phase space constructions, and path-integral formulations (Riello et al., 29 Sep 2025).

1. Quotients, constraint surfaces, and physical degrees of freedom

The basic structure of reduced phase space formalism is a symplectic manifold $(\mathcal{M},\omega)$ together with a symmetry action that preserves $\omega$. If the action is Hamiltonian with moment map $\mu$, the physical phase space is obtained by imposing the constraints $\mu=0$ and quotienting by the corresponding symmetry group. In the linear models used for abelian BF theory, this general scheme appears in both finite- and infinite-dimensional forms: for $V\oplus V^\ast$ one has the canonical linear symplectic form $\omega_0=\langle\delta v,\delta\alpha\rangle$, while for spaces of differential forms the pairing is given by integration over a hypersurface (Lysov, 2021).

The same structure appears in covariant field theory. One begins with an off-shell configuration space $\mathcal E$, an Euler–Lagrange locus $EL(L)\subset\mathcal E$, and a presymplectic current $\omega^h$ whose pullback to solutions is closed. The covariant phase space is then $(\mathcal{M},\omega)$0, which becomes presymplectic rather than symplectic in the presence of gauge symmetry. The reduced phase space is the quotient of this presymplectic space by the characteristic distribution generated by gauge transformations, or equivalently the quotient $(\mathcal{M},\omega)$1 when the gauge action is sufficiently regular (Riello et al., 29 Sep 2025).

A central implication is that “phase space” in gauge theory is not a single object. The data distinguish a bare or geometric phase space, a covariant phase space of solutions, a constraint surface inside the canonical phase space, a constraint-reduced symplectic space, and a fully reduced Poisson space. This suggests that reduced phase space formalism is best understood not as one quotient but as a hierarchy of related reductions, with symplectic and Poisson structures distributed across different stages (Riello et al., 29 Sep 2025).

2. Dirac reduction and extended phase space

A paradigmatic mechanical realization is obtained by promoting time to a canonical variable. For a time-dependent Hamiltonian

$(\mathcal{M},\omega)$2

the reparametrized action in an auxiliary parameter $(\mathcal{M},\omega)$3 leads to an extended phase space $(\mathcal{M},\omega)$4 with primary constraint

$(\mathcal{M},\omega)$5

This constraint is first-class and generates time reparametrizations; the reduced phase space is obtained by restricting to $(\mathcal{M},\omega)$6, quotienting by its gauge orbits, and choosing a time gauge such as $(\mathcal{M},\omega)$7 (Garcia-Chung et al., 2017).

After gauge fixing, the pair $(\mathcal{M},\omega)$8 becomes second-class, and the physical bracket is the Dirac bracket. In the simplest case discussed there, the only non-zero basic Dirac bracket is

$(\mathcal{M},\omega)$9

while $\omega$0 and $\omega$1 cease to be independent variables. The reduced phase space is then the submanifold coordinatized by $\omega$2, equipped with the canonical bracket and evolved by the original time-dependent Hamiltonian with respect to the physical time $\omega$3 (Garcia-Chung et al., 2017).

This formalism also clarifies the status of invariants. For the time-dependent harmonic oscillator,

$\omega$4

the Lewis invariant

$\omega$5

satisfies $\omega$6. It is therefore a Dirac observable: constant along gauge orbits and well defined on the reduced phase space. The same analysis shows that an extended canonical transformation can be canonical before reduction while failing to preserve Dirac brackets after reduction; such transformations are canonical on the enlarged constrained theory but not necessarily on the reduced physical phase space (Garcia-Chung et al., 2017).

The two-dimensional damped harmonic oscillator gives the same pattern in a regular system embedded into an extended singular one. After introducing $\omega$7 as a dynamical variable, the primary constraint

$\omega$8

becomes first-class; gauge fixing with a linear relation between $\omega$9 and the parameter $\mu$0 produces a second-class pair $\mu$1, and the reduced phase space is four-dimensional, with canonical Dirac brackets for $\mu$2 and $\mu$3. The physical Hamiltonian after reduction is precisely the original time-dependent Hamiltonian (Gouba, 2018).

3. Gauge theory, cohomology, and boundary reduction

In abelian BF theory without boundary, reduced phase space formalism becomes an explicit cohomological quotient. On a hypersurface $\mu$4, the bare phase space is

$\mu$5

Gauge transformations act by $\mu$6 and $\mu$7, with moment maps proportional to $\mu$8 and $\mu$9. Imposing the moment-map constraints sets $\mu=0$0 and $\mu=0$1 closed, and quotienting by exact shifts gives

$\mu=0$2

with reduced symplectic form inherited from $\mu=0$3 (Lysov, 2021).

When $\mu=0$4 has boundary, the naive cohomological quotient fails because the pairing becomes degenerate. The formalism is then extended by compact-support, relative, or mapping-cone constructions, together with edge modes. In the edge-mode picture one enlarges the phase space to

$\mu=0$5

and the symplectic form acquires an explicit boundary term. The corresponding moment maps acquire boundary components, their Poisson algebra becomes centrally extended, and the final invariant phase space is

$\mu=0$6

The reduction can be performed in stages, with finite-dimensional intermediate spaces describing edge modes or asymptotic symmetries (Lysov, 2021).

A closely related cohomological version appears in the BRST treatment of linear gauge theories on curved spacetime. There the classical phase space at ghost number $\mu=0$7 is defined as

$\mu=0$8

with hermitian form $\mu=0$9. This space is isomorphic to a quotient of space-compact solutions modulo BRST-exact directions, and it has a Cauchy-surface realization

$V\oplus V^\ast$0

Its non-degeneracy is governed by injectivity properties of the maps between smooth and compactly supported BRST cohomology. In the Maxwell case this reproduces the known de Rham criterion; in Yang–Mills, degeneracy can arise from local geometric data rather than non-trivial topology of the Cauchy surface (Wrochna et al., 2014).

4. Reduced phase space in general relativity

In canonical general relativity, reduced phase space methods are often tied to gauge choices or first-order formulations. In the radial gauge, one imposes

$V\oplus V^\ast$1

so that the radial coordinate is proper spatial distance from a distinguished point and the remaining canonical data are the intrinsic metric $V\oplus V^\ast$2 on the spheres $V\oplus V^\ast$3 and its conjugate momentum $V\oplus V^\ast$4. The diffeomorphism constraints are then solved by expressing the eliminated momenta $V\oplus V^\ast$5 as nonlocal functionals of $V\oplus V^\ast$6, $V\oplus V^\ast$7, and matter contributions. The reduced Poisson algebra remains canonical for $V\oplus V^\ast$8, while the Hamiltonian becomes nonlocal in the radial variable. This produces what the authors describe as an effectively $V\oplus V^\ast$9-dimensional geometry encoded by a one-parameter family of $\omega_0=\langle\delta v,\delta\alpha\rangle$0-geometries (Bodendorfer et al., 2015).

In Palatini–Cartan–Holst theory the reduced phase space is obtained covariantly from the boundary term in the variation of the action. The pre-boundary symplectic structure is degenerate along certain connection directions, and quotienting by this kernel yields a true boundary phase space that is canonically identified with a cotangent bundle over boundary coframes. Imposing the bulk equations gives Lorentz Gauss constraints and Einstein-type curvature constraints, whose Poisson algebra closes and defines a coisotropic submanifold. After reducing by internal Lorentz symmetry, one obtains the Einstein–Hilbert canonical variables $\omega_0=\langle\delta v,\delta\alpha\rangle$1, and the remaining constraints reduce to the standard ADM Hamiltonian and momentum constraints. Under the nondegeneracy assumptions stated in the paper, the reduced phase space of Palatini–Cartan–Holst theory is symplectomorphic to the reduced phase space of Einstein–Hilbert gravity. At the same time, the comparison with the “half-shell” version shows that bulk equivalence does not guarantee equivalence of boundary phase spaces (Cattaneo et al., 2017).

This combination of examples shows that, in gravity, reduced phase space formalism is sensitive both to gauge fixing and to boundary structure. It also shows that the same physical reduced phase space may arise from inequivalent unreduced or pre-boundary formulations, while apparently equivalent bulk theories may differ once boundary symplectic data are taken into account (Bodendorfer et al., 2015).

5. Quantization after reduction

Quantization after reduction does not take a single form. One possibility is finite-dimensional quantization of a compact reduced phase space. For the torus

$\omega_0=\langle\delta v,\delta\alpha\rangle$2

with $\omega_0=\langle\delta v,\delta\alpha\rangle$3, the Hilbert space is $\omega_0=\langle\delta v,\delta\alpha\rangle$4, the Weyl operators are built from a finite-dimensional representation $\omega_0=\langle\delta v,\delta\alpha\rangle$5 of the discrete Heisenberg group, and the Weyl map depends only on the values of the classical symbol on the lattice

$\omega_0=\langle\delta v,\delta\alpha\rangle$6

The map is non-injective: different classical symbols produce the same quantum operator, and equivalence classes are characterized by the map $\omega_0=\langle\delta v,\delta\alpha\rangle$7 acting on $\omega_0=\langle\delta v,\delta\alpha\rangle$8 samplings. The Wigner transform lives on a doubled lattice of $\omega_0=\langle\delta v,\delta\alpha\rangle$9 points, although only $\mathcal E$0 values are independent; the full support is required to maintain the correct marginals and normalization (Ligabò, 2014).

A second possibility is path integration on a deparametrized reduced phase space. In loop quantum gravity with Brown–Kuchař or Gaussian dust, the dust fields provide physical time and spatial labels, so the Hamiltonian and diffeomorphism constraints are solved classically and the remaining reduced phase space on a cubic lattice $\mathcal E$1 is coordinatized by holonomies $\mathcal E$2 and fluxes $\mathcal E$3, subject only to the SU(2) Gauss constraints. The physical Hamiltonian is

$\mathcal E$4

and the semiclassical stationary points of the corresponding coherent-state path integral satisfy

$\mathcal E$5

In the lattice continuum limit these equations reproduce the classical reduced phase space equations of gravity coupled to dust, provided the initial and final states are semiclassical (Han et al., 2020).

A third possibility is truncation of the quantum theory to finite-dimensional reduced phase spaces. In the quantum phase space trajectory approach, the quantum action is restricted to submanifolds of the Hilbert space parametrized by coherent-state labels $\mathcal E$6 together with additional fiducial variables $\mathcal E$7. The reduced action yields a finite-dimensional Hamiltonian system with symplectic form

$\mathcal E$8

after a suitable canonical transformation, and constraints given by normalization and physical centering conditions. Dirac’s method is then used to classify these constraints, and the resulting reduced trajectories provide approximations to full quantum evolution. In the cosmological example with reduced classical Hamiltonian $\mathcal E$9, the simplest truncation reproduces the familiar quantum bounce Hamiltonian $EL(L)\subset\mathcal E$0, while higher truncations encode nonclassical variables and dispersions (Małkiewicz et al., 2017).

6. Singular quotients, resolution, and asymptotic charges

A recurrent theme is that the reduced phase space need not be a smooth symplectic manifold. In gauge field theory the physical phase space typically emerges after a usually singular quotient with respect to the gauge group action, and the fully reduced space is often Poisson rather than symplectic. For manifolds with corners this reduction naturally separates into a symplectic constraint-reduced space $EL(L)\subset\mathcal E$1 and a further flux reduction to a Poisson space $EL(L)\subset\mathcal E$2, whose symplectic leaves are classical superselection sectors labeled by boundary fluxes (Riello et al., 29 Sep 2025).

This is one reason the literature distinguishes reduction from resolution. In the BV and BFV formalisms, one replaces explicit quotients by graded smooth objects carrying cohomological vector fields and master actions. Degree-zero cohomology reproduces functions on the reduced phase space, while the full graded structure resolves singularities, gauge identities, and boundary ambiguities without forming the quotient directly. The same framework makes explicit that on manifolds with boundary or corners, boundary terms, homotopies for the variational bicomplex, and Noether currents are part of the phase-space data rather than mere technical appendices (Riello et al., 29 Sep 2025).

Asymptotic and horizon phase spaces display the same pattern. In the covariant phase space analysis of Kerr spacetime in Bondi gauge, one studies diffeomorphism-generated perturbations $EL(L)\subset\mathcal E$3 and computes the corresponding Hamiltonian surface charges. Directions with vanishing Hamiltonian variation are pure gauge; directions with nonvanishing charges survive in the reduced phase space as physical soft directions. The resulting boundary charge algebra can be organized into a Virasoro algebra with central term, and the physical interpretation advanced there is that the reduced phase space consists of Kerr-like solutions modulo diffeomorphisms with vanishing charges, while the nontrivial surface-charge directions furnish the candidate microstate degrees of freedom (Setare et al., 2021).

A common misconception is that reduced phase space formalism always produces a unique smooth symplectic manifold of local observables. The examples surveyed here show otherwise. The quotient may be singular, only Poisson, or boundary-sensitive; canonical transformations may fail to remain canonical after reduction; and equivalence between bulk theories may fail at the level of boundary phase space. Reduced phase space formalism is therefore best understood as a family of constructions—Dirac reduction, symplectic reduction, cohomological quotient, and deparametrized Hamiltonian evolution—whose common aim is to isolate gauge-inequivalent degrees of freedom while preserving the correct physical bracket structure (Garcia-Chung et al., 2017).